Before you pay for an AI engineering course in Nepal, do three free things. Learn basic Python, call a free AI model from your own code, and finish the first unit of the Hugging Face Agents Course.

All three cost nothing. Together they tell you if a paid course will help you or bury you. Most guides to AI courses in Nepal compare degrees and entrance exams. Few say what you can do tonight on your own laptop.

Why start with Python, not with ChatGPT?

AI engineering is mostly normal programming around a model. You read a file, send text to a model, check the answer and save it. Every one of those steps is Python.

So the first free step is the official Python tutorial. You do not need all of it. Read chapter 5 on lists and dictionaries, chapter 7 up to section 7.2.2 on saving data as JSON, and chapter 8 on errors and exceptions.

Why those chapters? AI APIs send their answers back as JSON. Your code must read that JSON, and it must not crash when a reply comes back wrong. Those three chapters cover both.

Type every example yourself instead of copying it. A slow laptop is fine here. Python runs offline once it is installed, so a slow or shared connection will not stop you.

Can you use an AI API in Nepal for free?

Yes. Google AI Studio gives a free tier for the Gemini API. Its get-started page has four steps. Copy the API key AI Studio makes for you, save it in an environment variable called GEMINI_API_KEY, run pip install -U google-genai, then send one prompt from Python.

The page mentions billing only when you want higher rate limits. That matters in Nepal, where paying a foreign website usually needs a dollar card.

Build two habits from day one. First, never paste your key into your code or push it to GitHub. Keep it in the environment variable, or in a .env file that git ignores. Second, Google's pricing page says content sent on the free tier is used to improve its products. So send practice text only, never a customer's data or private files from your office or college.

Once one prompt works, change it. Ask the model to reply in JSON with two fields, then make your Python read those fields. If the reply breaks, catch the error instead of crashing. Every later AI project repeats that small loop.

Is the Hugging Face Agents Course worth doing first?

Yes, the first unit at least. The Hugging Face Agents Course is free. It asks for basic Python and a basic idea of what large language models (LLMs) are, which you now have.

Unit 1 covers how an agent works. The model thinks, picks a tool, acts, then reads the result. The unit ends with a small example that uses Python functions as tools. Finish Unit 1 and you get the free Fundamentals certificate, with no deadline.

Later units move into frameworks like smolagents, LangGraph and LlamaIndex. Leave those for now. They make more sense after you have built a few small things by hand.

Do you need a powerful laptop to start?

No. Everything above runs on a normal laptop, because the model runs on Google's servers, not on yours.

Running a model on your own laptop is a different story. Ollama makes it one command, ollama run gemma4. But the smallest Gemma 4 model in the Ollama library is listed at 4.6 GB to 7.5 GB. On an 8 GB laptop, that leaves little room for anything else. Wait on local models until you have a reason to need them.

What can free study not give you?

Free courses show you how to build things. What they cannot do is check your work.

A chatbot that answers ten test questions well can still fail on the eleventh. Finding that out needs a test set, a score, and a person who reads your code and tells you where it will break. That is where a mentor saves you weeks.

It is also where the AI Engineering and Machine Learning course at Saarathi Academy in Old Baneshwor, Kathmandu, spends real time. The course runs 12 weeks, Monday to Friday, two hours a day. There are up to 10 students in the classroom per batch, and a few online. Mentors still do the job they teach.

Python is taught from zero in Weeks 1 and 2. LLM work starts in Week 6, after machine learning basics. Week 10 is about evaluation: grading answers against a written rubric and guarding against prompt injection. The Hugging Face Agents Course is part of it too, as self-study in Week 9, so Unit 1 done now is a head start.

Class runs on the free tiers of Gemini, Groq, OpenRouter and Hugging Face. Paid OpenAI or Anthropic credit is optional.

How do you know you are ready to pay?

Try these three checks on your own, without following a tutorial line by line:

  • Write a Python script that reads a JSON file, loops over it and prints one field from each item.
  • Send a prompt to Gemini from that script, with the key kept out of your code.
  • Explain in your own words what an agent does that a plain chatbot does not.

If you can do all three, a fast course will feel hard but fair. Our course asks for 2 to 3 hours of practice every day on top of class, so plan your week for that. If the first check stops you, give Python two or three more weeks before you pay anyone.

What to try tonight

Open the AI Studio get-started page, copy your key and send your first prompt from Python. Then ask for the answer in JSON and print one field from it.

The AI Engineering and Machine Learning course page and the full syllabus show every week of the course before you pay anything.